TARI: Meeting Delay Requirements in VANETs

with Efficient Authentication and Revocation

Rex Chen Di Ma Amelia Regan
University of California, Irvine University of Michigan-Dearborn University of California, Irvine
rex@uci.edu dmadma@umich.edu aregan@uci.edu

Abstract - VANET based information systems have considerable participation of public agencies including national, state and
promise for improving traffic safety, reducing congestion and local transportation authorities such as the US Department of
increasing environmental efficiency of transportation systems.
Transportation (USDOT), the Federal Highway Association
However, the potential of these systems will not be realized until the
(an office of the USDOT), state departments of transportation
issue of network security is fully resolved. In this paper, we introduce
a security scheme called Temporary Authentication and Revocation or city and regional transportation agencies, which will
Indicator for VANETs. Using analytical and simulation based primarily be interested with improvements in public safety and
analyses, we examine the timing delay incurred by TARI and other congestion reduction. Further, it may well be that commercial
security metrics. We demonstrate that TARI is efficient enough to be
traffic will be unwelcome across networks delivering safety
deployed in safety critical VANET applications and can scale to dense
networks with many vehicles competing for broadcast air-time. and congestion critical messages to drivers in crowded traffic
networks.
I. INTRODUCTION Deploying successful large scale VANETs hinges on the
ability of these systems to absolutely guarantee the reliability
Traffic congestion and accidents continue to take a toll on
and security of all messages relayed. Our work focuses on
our society, with congestion impacts costing drivers enormous
both of these dimensions. In earlier research we examined the
amounts of time and accidents resulting in around 40,000
performance of broadcast communication and sought to
fatalities per year in each of the US and Europe [1] [2]. Further,
improve its reliability by explicitly incorporating fundamental
the number of fatalities each year in China alone is more than
relationships of traffic flow [4] to adjust the transmission
twice the number in the US and Europe because of its higher
power. Here we examine mechanisms for securing vehicular
population. Vehicle to vehicle (V2V), Vehicle to infrastructure
networks with message authentication. While many safety
(V2I), Infrastructure to vehicle (I2V) and Vehicle to pedestrian
applications exist, we focus on those in which communication
(V2P) communication systems have considerable promise for
is event driven rather than those in which it is periodic (such as
improving traffic safety, reducing congestion and increasing
cooperative adaptive cruise control or curve speed warning).
environmental efficiency of transportation systems. However,
We include single-hop broadcasting applications for both V2V
the potential of these systems will not be realized until the
and V2I communications. Representative applications of
issue of network security is fully resolved. Communication and
interest are shown in Table 1 [3] [5].
network security in Vehicular Ad hoc Networks (VANETs)
share some challenges in common: 1) Vehicles operate in a TABLE 1. REPRESENTATIVE SAFETY APPLICATIONS
dynamic network moving at fast speeds and may have very Safety Applications Communication Requirements
short connectivity time windows; and 2) communication and Comm Hop Latency Comm
security mechanisms need to be available even with an initial Type Traversal (msec) Range
(meters)
low market penetration rate of Dedicated Short Range Intersection Collision Single-
Communication (DSRC) enabled inter-vehicle communication. V2V ~100 ≤ 300
Warning hop
A review of the primary VANET applications is provided in Single-
Pre-Crash Sensing V2V ~20 ≤ 50
hop
[3]. These are classified as safety, convenience and commercial
Transit Vehicle Signal Single-
applications. Our interest lies primarily in safety applications, V2V ~100 ≤ 1000
Priority hop
with a secondary interest in those providing convenience (note Approaching Emergency
V2V
Single-
≤ 1000
~1000
that this category includes real-time traffic information in Vehicle Warning hop
addition to services such as parking availability and location
notification). While commercial applications may indeed Message authentication is one of the most important
provide a tertiary benefit to automobile drivers, it seems likely requirements for VANET security. Digital signature is the
that by the time VANETs are fully functional, most of these most widely accepted solution for message authentication [6].
applications will already be fairly mature and will rely on Each message is digitally signed before transmission and its
cellular communication systems or other wireless technologies. signature is verified before the message is taken under
In addition, deploying functional VANETs will require the consideration. Since safety messages are sent out periodically
by participating vehicles, in a dense network, a vehicle’s on-
board unit (OBU) may receive hundreds of such messages in a
short time span. The ability to verify these digitally signed
messages quickly presents some challenges for OBUs since in
order to keep the cost low, OBUs have limited computation
power. A typical OBU has a 400MHz CPU [7].
If an OBU cannot verify all these messages, some will not
be processed. These un-processed messages may contain
urgent and safety critical information. Lack of knowledge may
contribute to accidents, and even death. Thus, for a security
solution to be practical and useful, it is critical to consider
entity capabilities and to develop a design suitable for those
applications.
In this paper, we propose TARI (Temporary Authentication
and Revocation Indicator) for efficient authentication and
revocation while meeting delay requirements in VANETs. In
TARI, a symmetric key based approach is used for V2V
communication for efficiency in both computation and
communication. The TARI authentication is based on TESLA
broadcast authentication [8] while revocation is based on a
revocation mechanism in the Certificate Revocation System [9].
II. RELATED WORKS
Several recent papers examine security challenges in
VANETs and discuss basic mitigation measures [6] [7] [10]
[11] [14] [15]. Securing V2V communication requires several
conflicting security properties:

Authenticity: to provide entity and message
authenticity.

Privacy: to provide long-term anonymity and prevent
tracking.

Traceability and revocation: to trace and disable
abusing OBUs in a timely manner.
Multiple schemes have been proposed to achieve both
authentication and anonymity. The simplest way would be to
equip each vehicle with a large set of certified public/private
keys during the process of yearly security inspections [11].
Each key would be used only for a brief period of time.
However, this approach is subject to Sybil attacks and would
also require that each vehicle be inspected each year. Another
is to use group signatures within VANETS [12]. However,
group signatures are computationally expensive, making
frequent use of group signatures infeasible to meet delay
requirement in safety VANET applications.
Recently, Studer et al. proposed the TACK scheme [13]. In
TACK, Regional Authorities (RA) are used to issue temporary
public/private keys. The road is divided into multiple regions
and when an OBU enters a region, it sends an authenticated
request to the RA for a temporary anonymous certified key
(TACK). An OBU is equipped with a group user key and the
request is authenticated through a group signature generated by
OBU’s group user key. A group signature allows the OBU to
authenticate anonymously. An OBU uses TACK for
subsequent V2V communication. TACK satisfies all the above
mentioned security requirements. Since subsequent V2V
communication uses the same temporary key for authentication,
the TACK scheme also provides short-term linkability, a
desirable property in some applications [13].
Other recent works [14] and [15] discuss the extensive
findings of the EU Secure Vehicle Communication (SeVeCom)
project.
III. SYSTEM MODEL
We consider the following entities in TARI:

MA: managing authority or root of trust. It certifies
OBUs and RAs. A MA can consist of multiple entities
that include a Department of Motor Vehicles (DMV) or
other national, regional, state and local transportation
authorities. MA works off-line most of the time and
certifies both RA and OBU.

RA: regional authority. A RA is part of the road-side
infrastructure. It stays online to process OBU requests
and issues temporary keys for V2V communication in
the region. We assume the road is divided into multiple
regions covered by different RAs. An OBU can contact
at least one RA when entering a region. A RA is
certified by the MA.

V: set of legitimate OBUs that have a valid certificate
from the MA, and can use the valid certificate to get a
TARI from RA.

V’: revoked OBUs which are included in the MA's
revocation list and do not have a valid TARI.
A. Goals
The goals of TARI are as follows:

Authentication. TARI should support multiple
authentications including OBU to RA, RA to OBU and
OBU to OBU authentication.

Privacy. RAs or other vehicles should not be able to
track a participating OBU on the long term.

Traceability and revocation. An abusing OBU should be
identified and its ability in participating VANETS
should be disabled.

Efficiency. Cryptography used in TARI should be
computational efficient since OBUs have resource-
limited processors.
B. Assumptions
We assume a passive adversary who listens to all
communication traffic. The adversary’s goal is to forge a valid
signature/message authentication code (MAC) and impersonate
a valid entity to participate in VANET communication. We do
not consider a compromised vehicle that is under full control of
the adversary. Further, we do not consider denial-of-service
attacks as it is difficult to prevent an adversary from jamming
the wireless channel.
We assume loose synchronization among vehicles in the
same region. By loose synchronization, we mean that a
receiver does not need to know exact clock differences
between itself and the sender. It only needs to know the upper
bound of the differences. Security of TESLA-based
authentication is based on this loose synchronization. Many
schemes, including the one proposed in the original TESLA
paper [8] can be used to synchronize clocks of neighboring
vehicles in a region.
The notation used in the remaining of the paper is
summarized in Table 2.
TABLE 2. NOTATIONS
Symbol Definition
gSign Group signature generation
gVerigy Group signature verification
sign (normal) signature generation
verify (normal) signature verification
Enc Public key encryption scheme
Dec Public key decryption scheme
gpk Group public key
guk Group user private key or OBU’s
private key
pkRA RA public key
prRA RA private key
IV. DESIGN
Our design of TARI follows the structure of TACK: we use
a group signature scheme to provide anonymity as well as to
allow tracing and revocation. Each valid OBU has a unique
group user key guk issued by a trusted MA. The guk is stored
on the OBU and remains there for a reasonably long period of
time. The MA records all issued group user keys. Once an
OBU misbehaves, the MA can trace the identity of the
misbehaving OBU and revoke its guk.
When an OBU enters a region, it sends a request to the
regional authority RA for a temporary TARI. The request
contains a group signature generated by the OBU’s guk. After
the RA successfully verifies the request, it issues a short-lived
TARI to the OBU. The TARI is used for both authentication and
revocation. In the following, we show how the TARI is
requested and generated and discuss how it can be used for
message authentication and OBU revocation.
C. TARI Request and Generation
A TARI consists of two components: an authentication
indicator (AI) and a revocation indicator (RI). The AI is used by
the destination OBU to verify a message from a source OBU
while the RI is used to disable the corresponding AI. After
revocation, AI cannot be used to authenticate messages
generated by the source OBU.
When an OBU enters a region, it requests a TARI from the
RA in that region. The OBU first chooses two 8-byte random
numbers Y0 and N0. Then it encrypts the two values with the
RA's public key. Finally, the OBU generates a group signature
over the encrypted AI and RI and sends them to the RA to
request a TARI.
After the RA successfully verifies the OBU’s request, it
extracts Y0 and N0 from the received message by decrypting
with its private key prRA. It then computes AI = Hmax(Y0) and RI
= H(N0) where H() is a pseudo-random function (PRF) and
Hmax() denotes repeatedly applying the function H() over the
inputs max times. A TARI is a certificate issued by the RA on
AI and RI. max is a system parameter determined by the RA
that indicates the maximum number of time intervals a TARI
will remain valid. Suppose an OBU broadcasts a safety
message every 100 ms and a TARI is expected to be valid for
two minutes, we can set max = 1200 (2 min/100 ms). The RA
logs all requests and issue TARIs for future tracing and
revocation.
The TARI request and generation steps are shown in Table 3.
TABLE 3. TARI REQUEST AND GENERATION
N0 ←  $ : Y0 ← $
R R
OBU:
OBU: Compute c = EncpkRA(N0, Y0)
σ = gSignguk (c)
OBU → RA: c, σ
RA: Use gVerify to verify σ
If verification is successful extract N0,Y0 from
c by decrypting c using PrRA
RA: Compute RI = H (N0) and AI = Hmax (Y0)
where max is a system parameter chosen by
RA. H(·) is a cryptographic hash function
RA: Compute σ' = SignprRA (AI, RI) and TARI =
(AI, RI, σ')
RA → OBU: TARI
D. V2V Communication with TARI
Each OBU broadcasts its valid TARI periodically to the
network. Since a TARI is certified by the RA and we assume
OBUs within the region know the RA’s public key, each TARI
can be verified by OBUs in the region. After successfully
verifying a TARI, an OBU stores AI and RI for subsequent
communication.
We divide time into transmission time periods and assume
that a single safety message is generated during each
transmission time period (100 ms). A safety message msgi at
time i has a format (AI, msgi, maci, Ki-1) where maci is the
Message Authentication Code of msgi under a symmetric key
Ki.. That is: maci= MACKi(msg). An OBU generates its i-th
period key Ki as Ki = Hmax-i(Y0).
TARI delays message verification by one transmission
interval (100 ms). After an OBU receives a broadcast message
(AI, msgi, maci, Ki-1), it stores the message after first making
sure that AI is in its verified list and then uses Ki-1 to verify
msgi-1 received in the previous transmission interval.
The verification takes two steps. First, the OBU verifies Ki-1
by comparing whether Hi-1(Ki-1) equals AI. If the two values are
equal, Ki-1 is authenticated and the OBU can use it to verify
msgi-1. Since an OBU can store the latest key after it is verified,
the first step in the verification process needs only one hash
function invocation.
E. Revocation with TARI
When an OBU is found to be misbehaving, revoking its
long-term group user key is done through the revocation
process of the group signature scheme. In this section, we
discuss how to revoke a TARI and the corresponding AI.
Revoking a TARI is simple and efficient. RA broadcasts a
revocation message in the form of (AI, N0). When an OBU
receives the revocation message, it first checks whether AI is in
its record list and then compares N and H(N0). A match
indicates the corresponding AI or TARI is revoked. So the OBU
removes the pair (AI, N) from its record list and refuses to
accept messages authenticated with AI. Since the source OBU
knows N0, it can also broadcast a revocation message (AI, N0)
to do self-revocation. Self-revocation is performed when the
source OBU wants to reduce the short-term linkability period.
V. PERFORMANCE ANALYSIS
In this section, we first discuss how TARI meet the
necessary requirements for security analysis analytically and
communication through simulation. Then, we discuss the
timing analysis and how TARI meet delay constraints for
safety critical VANET applications.
A. Security Analysis
Sender authentication. TARI authentication is based on the
existence of a vehicular PKI with MA as the root of trust. Both
RAs and OBUs are certified. All authentications, OBU to RA,
RA to OBU, OBU to OBU, are based on the trust of the MA.
Message integrity. The security of TESLA-based message
authentication relies on loose synchronization between sender
and receiver. We assume a TESLA interval is smaller than a
transmission time interval (100 ms). We can set a TESLA
interval as the upper bound for VANET propagation delay (for
example 10ms). Hence, a transmission interval may consist of
multiple TESLA intervals. However unlike the original
TESLA scheme, only the first TESLA interval is associated
with a TESLA key. A key is disclosed after multiple TESLA
intervals. In this way, the adversary cannot forge a valid MAC
before the key is disclosed. After the key is disclosed the
adversary can forge a valid MAC. However, an OBU will find
the message to be malicious and discard it.
Long-term unlinkability. Long-term unlinkability is
achieved partly using group signatures, and partly through
short-lived TARIs. Only the MA can distinguish group
signatures generated by different OBUs. So a group signature
provides anonymity. Even though TARI provides short-term
linkability, TARI is short-lived. When multiple OBUs change
TARIs at one time – which happens often in dense networks,
an adversary will not be able to link an old TARI with a new
TARI.
Short-term linkability. Before expiration of a TARI, all
messages are authenticated through AI which can be thought of
as a temporary ID.
Traceability and revocation. Both MA and RAs keep
records of all issued certificates. Whenever abusing events
happen, MA and RA collaborate to identify abusing vehicles.
MA can distinguish group signatures generated by different
group users. To revoke an abusing OBU’s long term group
user key, MA will issue an updated RL containing the revoked
OBU. Since N0 is only known to RA and OBU itself, it can
revoke an abusing OBU.
The estimated computational time for certificate creation,
signature generation and verification time are described in
Table 4 and Table 5 and was previously investigated by [7] and
[13].
TABLE 4. ESTIMATED TIME IN GROUP SIGNATURE AND CERTIFICATE
CREATION
Operation
Computational Data Size
Time (ms) (bytes)
OBU Group Sig Creation 320 228
RA Group Sig Verify 36 228
RA Creation of Certificate 3.2 28
TABLE 5. COMPARISON OF SIGNATURE GENERATION AND VERIFICATION TIME
Sig Size (bytes) Generation (ms) Verification (ms)
ECDSA 28 3.255 7.617
NTRU 197 1.587 1.488
TARI 8 0.001 0.002
B. Communication Analysis
We use the ns-2.33 network simulator to analyze the
communication performance. Different security mechanisms
have varying byte overhead. We specifically look at the end-to-
end delay metric incurred in the wireless communication with
security bytes embedded in the packet data. To have a holistic
perspective of security delay, we examine the communication
performance with varying packet sizes with 1000 meter node
distance and a message generation rate of 100 ms intervals.
The packet data include the safety application payload and
security overhead (bytes from message signature and
certificate). For higher fidelity in the ns-2 simulation, we use
IEEE 802.11p configuration values and run an entire day of
simulation time to reduce variance in the results. The preferred
data rate of 6 Mbps for vehicular safety applications is used in
our simulation, which has the greatest benefit in overall
reliability (in terms of packet reception rates) as confirmed by
[16].
The effect of security schemes and application data size on
VANET performance is shown in Figure 1. First, delay
increases linearly with larger packet sizes, a difference of 0.6
ms from 50 to 500 bytes. Also, note that NTRU has
significantly larger security overhead (591 bytes) and hence its
delay is on the order of 0.75 ms longer than ECDSA and TARI.
At the communication stack lower layers, several factors can
impact delay (for example the inter-frame spacing and slot
times in the medium access control layer). A major contributor
of delay in medium access control is the back-off time,
calculated based on the contention window size.

Figure 1. Packet Data with Security PKCS
C. Delay Analysis
To understand whether a secure message meets the safety
application requirements, we compute delay that includes
security processing and communication transmission time. A
data rate of 6 Mbps is used along with a 200 byte data payload
(a realistic upper bound according to [5].
The total delay for the certificate request (V2I) and
certificate retrieval (I2V) includes the following steps:
1. OBUsrc group signature creation
2. OBUsrc send TARI/TACK request to RA
3. RA group signature verification
4. RA create TARI/TACK certificate (ECDSA)
5. RA send certificate back to OBUsrc
6. OBUsrc verify certificate (ECDSA)
The total delay for delivering data messages between
vehicles (V2V) includes the following steps:
1. OBUsrc generate signature for msg
2. OBUsrc send msg to OBUdest
3. OBUsrc verify msg
Using numerical values from sections 5-A and 5-B, we
compare the aggregate delay (based on the formulations above)
for TACK - with ECDSA for V2V communication - and TARI
in Table 6. As noted earlier, TARI and TACK have the same
operation for certificate generation and verification. Its
numerical values on timing for the V2I and I2V relay are the
same. The major benefit of TARI is realized when delivering
messages between vehicles. In TARI, security delay from the
required hash invocations is trivial, compared to the
communication delay. It is generally believed that at least half
a second is needed for a driver to respond in urgent situations.
Although TARI delays the verification of messages by one
transmission interval due to the hashing scheme, it still satisfies
this response time.
TABLE 6. OVERALL DELAY COMPARISON
V2I + I2V (ms) V2V (ms)
TACK 368.0 11.4
TARI 368.0 0.6
Further, we note that certificate generation and verification
occurs only when a vehicle contacts the RA for a certificate
which occurs when the vehicle initially enters the RA coverage
area or when the certificate life-time expires and a new
certificate request is necessary. The certificate life-time
impacts short-time linkability and the certificate revocation
process. Although a shorter life-time may be ideal from a
security standpoint, each certificate request results in a
computation cost of 368 ms and additional requests can
decrease communication reliability because more messages are
competing for air-time. In time-critical safety applications this
is undesirable. In TACK, the certificate life-time is both timer
(typically set to 5 to 20 minutes) and region-specific.
We additionally suggest that an optimal expiration life-time
should be responsive to the prevailing traffic flow
characteristics. For example, heavy peak hour traffic results in
high network density. A pro-longed certificate life-time can
alleviate wireless interference and reduce packet loss due to
additional certificate requests. When a shorter certificate life-
time is preferred, having RAs placed at strategic locations can
be beneficial to compensate for the delay incurred as vehicles
are more likely to slow down or stop at these locations (for
example at the ramp meters already found on major highway
on-ramps).
D. Discussion
Based on the numerical results in Table 5 and the application
requirements shown in Table 1, both TACK and TARI satisfy
the minimal 100 ms latency for safety applications under the
most ideal traffic scenarios with no interference from other
transmissions. We consider two scenarios envisioning a
realistic traffic topology and reasonable communication values
but under the worst case condition when all vehicles contend
for the wireless channel. For the traffic topology, we assume a
highway system with 6 lanes and traffic which is spatially
uniformly distributed. Numerical values for the traffic network
are derived according to the speed-density relationship based
on the triangular fundamental diagram [17] [18] and by
assuming free flow speed Vf = 104 km/h (64.6 mph), jam
density ρj = 150 veh/km (240 veh/mi), and critical density ρ c =
0.2 ρj. The safety data payload is fixed at 200 bytes.
Scenario 1: Free-Flow Traffic on Highways
The critical density (30 veh/km) is selected for free-flow
traffic with the most vehicles, with vehicle spacing of 33.3
meters. The transmission range is 250 meters and messages are
generated at 100 ms intervals. When all vehicles are trying to
send messages, the system throughput required by TACK and
TARI are 1.91 Mbps (84 veh x 10 msg/sec/veh x 284
bytes/msg) and 1.51 Mbps (84 veh x 10 msg/sec/veh x 224
bytes/msg). For authentication, the maximum tolerable [4] R. Chen, W. Jin, and A. Regan, "Multi-Hop Broadcasting in
processing delay to verify each message is 100 ms/84= 1.19 ms.
Scenario 2: Congested Traffic on Highways
A congested density (90 veh/km) is selected to mimic a stop-
and-go traffic (assuming an average speed of 10.8 mph) on [6] M. Raya and J. Hubaux, “Securing Vehcular Ad Hoc Networks”,
highways with vehicle spacing of 11.1 meters. The
transmission range is 100 meters (to reduce communication
density for higher packet reception rates) and messages are
generated at 100 ms intervals. When all vehicles are trying to [8] A. Perrig, R. Canetti, J. Tygar, and D. Song, "The TESLA
send a message, the system throughput required by TACK and
TARI are 2.45 Mbps (108 veh x 10 msg/sec/veh x 284
bytes/msg) and 1.94 Mbps (108 veh x 10 msg/sec/veh x 224 [10] B. Parno and A. Perrig, "Challenges in securing vehicular
bytes/msg). For authentication, the maximum tolerable
processing delay to verify each message is 100 ms/108= 0.93 [11] M. Raya, P. Papadimitratos, and J. Hubaux, "Securing Vehicular
ms.
In both scenarios, the system throughput is below the
minimal DSRC data rate of 3 Mbps. No communication
capacity issue is seen here. However, the benefit of using
TARI to verify messages is important because TARI is able to
meet the maximum tolerable processing delays, 1.19 ms and
0.93 ms. As shown earlier in Table 5, ECDSA and NTRU
cannot meet these delay constraints. Further, TARI can handle
far more messages than TACK – resulting in more
authenticated messages.
VI. CONCLUSION AND FUTURE WORK
In this paper, we show that our proposed security scheme
TARI is efficient enough to be deployed in safety critical
VANET applications. Further, we show that it can scale to
dense networks with many vehicles competing for broadcast
air-time.
Our ongoing research seeks to identify and solve
communication reliability and security issues in VANETs
while incorporating an understanding of the underlying traffic
flow dynamics and the impact of traffic flow on the temporal
and spatial configuration of communicating vehicles in the
network. Further, we plan to explore security analysis with a
thorough comparison of all representative signature schemes
with the application requirements for VANETs.
Vehicular Ad Hoc Networks with Shockwave Traffic," IEEE
Intelligent Vehicular Communications System Workshop, 2010.
[5] CAMP VSC, "Vehicle Safety Communications Project Task 3
Final Report - Identify Intelligent Vehicle Safety Applications
Enabled by DSRC," USDOT, 2005.
Journal of Computer Security, 2007.
[7] M. Raya and J. Hubaux, "The Security of Vehicular Ad Hoc
Networks," ACM workshop on Security of Ad Hoc and Sensor
Networks, 2005.
broadcast authentication protocol," RSA CryptoBytes, 2002.
[9] S. Micali, "Enhanced certificate revocation system," Technical
Report (MIT/LCS/TM-542), 1995.
networks," ACM Workshop on Hot Topics in Networks (HotNets-
IV), 2005.
Communications," IEEE Wireless Communications, 2006.
[12] J. Guo, J. Baugh, and S. Wang, "A group signature based secure
and privacy-preserving vehicular communication framework,"
Mobile Networking for Vehicular Environments (MOVE)
workshop in conjunction with IEEE INFOCOM, 2007.
[13] A. Studer, E. Shi, F. Bai, and A. Perrig, "TACKing Together
Efficient Authentication Revocation, and Privacy in VANETs,"
IEEE Communications Society Conference on Sensor, Mesh and
Ad Hoc Communications and Networks (SECON), 2009.
[14] P. Papadimitratos, L. Buttyan, T. Holczer, E. Schoch, J.
Freudiger, M. Raya, Z. Ma, F. Kargl, A. Kung, and J. Hubaux,
"Secure vehicular communication systems: design and
architecture," IEEE Communications Magazine, 2008.
[15] P. Papadimitratos, L. Buttyan, T. Holczer, E. Schoch, J.
Freudiger, M. Raya, Z. Ma, F. Kargl, A. Kung, and J. Hubaux,
"Secure vehicular communication systems: design and
architecture," IEEE Communications Magazine, 2008.
[16] D. Jiang, Q. Chen, and L. Delgrossi, "Optimal Data Rate
Selection for Vehicle Safety Communications," ACM
International Workshop on Vehicular Ad Hoc Networks
(VANET), 2008.
[17] F. Hall, B. Allen, and M. Gunter, "Empirical analysis of freeway
flow-density relationships," Transportation Research Part A, vol.
20, 1986.
[18] R. Munjal and Y. Hsu, "Analysis and validation of lane-drop
effects of multilane freeways," Transportation Research, 1971.

VII. ACKNOWLEDGEMENT

This research was supported in part by the University of

California Transportation Center.

REFERENCES
[1] NHTSA, "Traffic Safety Facts," Technical Report, 2009.
[2] European Commission, "Directive of the European Parliament and
of the Council - facilitating cross-border enforcement in the field
of road safety," Technical Report, 2008.
[3] F. Bai, T. Elbatt, and G. Hollan, "Towards characterizing and
classifying communication-based automotive applications from a
wireless networking perspective," IEEE Workshop on
Automotive Networking and Applications (AutoNet), 2006.